Research On Prediction Algorithm Of Missing Link In Complex Network

Link prediction has been the focus of web data mining and prediction.For the reasons that the topology information of the network is more easily accessible and reliable,the missing link prediction algorithm based on the similarity of the network topology has become a research hotspot.However,due to the complexity and diversity of real networks,the current algorithms of missing link prediction cannot have good prediction accuracy in most networks.Through analyzing the link formation in real networks and the process of dynamic changing,this paper proposes a missing link prediction algorithm named MJR-FCN on the basis of future common neighbor and short path topology to improve the universality of the previous algorithm.And we verify the performance of the algorithm by experimental simulation.The main works of this paper are as follows:Firstly,considering the physical evolution of the links and the dynamic changes of the network,the concept of future common neighbors is introduced to solve the limited utilization of nodes information.Besides,the number of common neighbors and the celebrity effect are used to measure the contribution of each node that may be a common neighbor in the network,and the Future Common Neighbor(FCN)is proposed.Then,from the perspective of the network topology,the future common neighbor only considers the role of the first-order and second-order paths.To extract more effective information from the network nodes,this paper considers the role of the number of first-order,second-order and third-order paths,and proposes a lost link prediction index MJR based on short-path topology.Secondly,since that FCN and MJR are the missing link prediction indexes proposed from different perspectives,to obtain higher prediction accuracy in networks with different characteristics,this paper makes full use of the factors that contribute to missing link predictions and proposes a mixed algorithm of missing link prediction named MJR-FCN that is based on future common neighbor and short path topology with adaptive parameters.Finally,in order to verify the rationality of the algorithm proposed in this paper,we use three algorithms(FCN,MJR and MJR-FCN)respectively to do 100 independent experiments on datasets of eight networks.The experimental results show that the prediction accuracy of FCN and MJR indicators on the eight networks has been improved,and the two indexes are complementary.The MJR-FCN indicators utilize this feature to achieve better universality for missing link predictions and higher forecasting accuracy.Particularly,the prediction accuracy of the Power network and the FWFB network has been significantly improved by 24% and 20%,respectively.